Session: 06-08-01 Model Tests - I

Submission Number: 176021

Experimental Evaluation of Data-Driven Deterministic Wave Generation 

The controlled generation of tailored wave sequences in laboratory environments is essential for advancing research in hydrodynamics, coastal engineering, and renewable energy systems. While simple models are sufficient for generating stochastic and weakly nonlinear deterministic sea states, their accuracy diminishes with increased wave steepness. Conventional practice therefore resorts to iterative optimizations to the wavemaker motion, a workflow that is timeintensive and thus impractical for many commercial applications. The data-driven identification of an experimental wave tank setup bypasses the need for explicit hydrodynamic modeling by learning the complex relations between wavemaker dynamics, wave surface dynamics, and wave basin response from measurement data. By learning a direct mapping between the design wave and the wavemaker control signal, data-driven methods enable the accurate generation of deterministic wave sequences.

In this work, a fully convolutional neural network (FCNN) is deployed to generate tailored waves in an experimental setup. The FCNN architecture is trained to map the wave sequence at a target location to the wavemaker control signal. The same architecture is fit to a variety of training datasets that comprise different combinations of experimental- and synthetic data. The resulting models are compared against a classical approach based on analytical wavemaker transfer functions (linear) and linear wave theory (LWT) to propagate the wave from the wavemaker to the target location. All methods are validated experimentally by generating short deterministic wave sequences that are compared to the predefined target wave at target location. The experimental evaluation is conducted on 950 irregular waves within a wide range of sea state parameters spanned by varying JONSWAP peak enhancement factors, angular peak frequencies and wave steepness. Furthermore, a set of 14 tailored extreme waves, including the New Year Wave, is used to evaluate the methods on the generation of extreme events. It is shown that the data-driven approach consistently outperforms the classical method in accuracy across the studied range of sea states, emphasizing its effectiveness as a flexible and precise solution for wave basin experiments.

Presenting Author: Marco Klein German Aerospace Center (DLR), Institute of Maritime Technologies and Propulsion Systems

Presenting Author Biography: Dr.-Ing. Marco Klein graduated as Diplom-Ingenieur (Dipl.-Ing.) in naval architecture and ocean engineering at the Technical University Berlin (TUB) in 2006. Prior to his current position as Head of Ship Performance Department at DLR Institute of Maritime Technologies and Propulsion Systems (formerly: Institute of Maritime Energy Systems), he worked as senior researcher at Institute for Structural Dynamics Hamburg University of Technology (TUHH), as post-doctoral researcher at Institute for Ship Structural Design and Analysis (TUHH) as well as research assistant at the Ocean Engineering Division at TUB where he received his doctorate in 2015. He received the Curt-Bartsch-Award (2015) from the German Society of Naval Architects (STG) for his doctoral thesis „Tailoring critical wave sequences for response based design“.

Authors:

Mathies Wedler German Aerospace Center (DLR), Institute of Maritime Technologies and Propulsion Systems
Marc-André Pick Hamburg University of Technology - Institute of Mechanics and Ocean Engineering (M-13)
Robert Seifried Hamburg University of Technology, Institute of Mechanics and Ocean Engineering (M-13)
Norbert Hoffmann Hamburg University of Technology, Dynamics Group (M-14)
Sören Ehlers German Aerospace Center (DLR), Institute of Maritime Technologies and Propulsion Systems
Marco Klein German Aerospace Center (DLR), Institute of Maritime Technologies and Propulsion Systems